Design of experiments and progressively sequenced regression are combined to achieve minimum data sample size

Abstract

The number of samples needed to create mathematical models is an issue as manufacturers see benefits to customers. Testing to acquire scores or hundreds of samples in a production environment is costly. This paper explains a method for verifying minimum sample size. Latin hypercube experiment design strategy (LHC sampling) is used along with a new method, progressively sequenced regression (PSR) analysis. The number of samples becomes, essentially, an independent variable, which the user has control over. PSR analysis provides efficacy verification. Specific steps must be followed to prevent false positives and negatives, but no new technology is used. This paper presents the principles, the procedures that are required, and shows empirical flow model results using a typical variable displacement piston pump. It is an empirical paper and is presented as an introduction to a technique that has the potential for other uses.